Every LLM Request Has Two Halves. Only One Uses Your GPU Cores

Every LLM request has two halves that behave nothing alike. One explains the pause. The other explains the typing.